ANNALES DE L INSTITUT FOURIER

Transcription

1 ANNALES DE L INSTITUT FOURIER YUM-TONG SIU An effective Matsusaka big theorem Annales de l institut Fourier, tome 43, n o 5 (1993), p < 43_5_1387_0> Annales de l institut Fourier, 1993, tous droits réservés. L accès aux archives de la revue «Annales de l institut Fourier» ( implique l accord avec les conditions générales d utilisation ( Toute utilisation commerciale ou impression systématique est constitutive d une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright. Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques

2 Ann. Inst. Fourier, Grenoble 43, 5 (1993), AN EFFECTIVE MATSUSAKA BIG THEOREM by Yum-Tong SIU (*) Dedicated to Professor Bernard Malgrange 0. Introduction. In this paper we will prove an effective form of the following Matsusaka Big Theorem ([Ml, M2, LM]). Let P(k) be a polynomial whose coefficients are rational numbers and whose values are integers at integral values of k. Then there is a positive integer ko depending on P(k) such that, for every compact projective algebraic manifold X of complex dimension n and every ample line bundle L over X with f^ (-l^dim^x.^l) == P(k) v=q for every fc, the line bundle kl is very ample for k > ko. Here ampleness means that the holomorphic line bundle admits a smooth Hermitian metric whose curvature form is positive definite everywhere. Very ampleness means that global holomorphic sections separate points and give local homogeneous coordinates at every point. By a result of Kollar and Matsusaka on Riemann-Roch type inequalities [KM], the positive integer ko can be made to depend only on the coefficients of ^ and A: 71 " 1 in the polynomial P(k) of degree n. The known proofs of Matsusaka's Big Theorem depend on the boundedness of numbers calculated for some varieties and (*) Partially supported by a grant from the National Science Foundation. Key-words : Ample line bundle - Very ample line bundle - Positive line bundle - Matsusaka Big Theorem - Holomorphic line bundle - Compact complex manifold - d" estimates - Closed positive current - Leiong number - Strong Morse inequality. A.M.S. Classification : 32J25-14E25-14C30.

3 1388 YUM-TONG SIU divisors in a bounded family and thus the positive integer ko from such proofs cannot be effectively computed from P(fc). To state our effective Matsusaka Big Theorem, we use the following standard notation. For holomorphic line bundles Li,---,Z^ over a compact complex manifold X of complex dimension n with Chern classes c(i/i),,c(i/^) and for positive integers k\,, kj^ with k\ -{- -h k(, = n, we denote the Chern number 0(2.1)^. 'c(l^ by L^'-L^. Our effective version of Matsusaka's Big Theorem is the following : THEOREM (0.1). Let L be an ample holomorphic line bundle over a compact complex manifold X of complex dimension n with canonical line bundle Kx- Then ml is very ample for (2 3n " 1 5yl) 4T^" l (3(3n - 2) n L n + Kx ' L 71-1 ) 471 " 13^ m ~ (6(3n - 2) 71-2n - 2) 4T^-ln -j(l 7l ) 4Tl - 13 ( n - l ) Theorem (0.1) is a consequence of the following Theorem (0.2) when the numerically effective holomorphic line bundle B in Theorem (0.2) is specialized to the trivial line bundle. Here the numerical effectiveness of B means that the value of the first Chern class of the line bundle B evaluated at any complex curve is nonnegative. THEOREM (0.2). Let X be a compact complex manifold of complex dimension n and L be an ample line bundle over X and B be a numerically effective holomorphic line bundle over X. Let H = 2(Kx + 3(3n - 2) n L). Then ml - B is very ample for (^(Jr 1 ) 2^71-1. B + JfT 1 )) 471 " 1 m > (6(3n - 2) 71-2yi - 2) 4n ~ ln -j The result of D ly [D2] that 12 n^l + 2 Kx is very ample for any ample line bundle L over X can also be regarded as an effective version of Matsusaka's Big Theorem. The difference between D ly^s result and Theorem (0.2) is that in Theorem (0.2) 2Kx is no longer needed and in its place B can be used for any ample line bundle B. On the other hand the coefficient 12 n^ of L in D ly's result depends only on n and is far sharper, whereas in Theorem (0.2) the coefficient for L depends on L 71, L 71-1 Kx, L 71-1 ' B, and I"- 2 B ' Kx as is expected. (The condition in Theorem (0.2) is expressible in terms of L 71, L Kx, ^n-l B, and L 71 " 2 B ' Kx by using inequalities of Chern numbers of numerically effective line bundles (see [D2, Prop. 5.2(b)] and the end of 4.) Kollar

4 EFFECTIVE MATUSAKA THEOREM 1389 [K] gave an algebraic geometric proof of a result similar to but weaker than D ly's criterion for very ampleness. For example, Kollar's result gives the very ampleness of (n + 3)! 2(n + 1) ((n + 2)L + Kx) + Kx for an ample line bundle L over X (by replacing L by (n + 2)L + Kx and setting a = 1 in Theorem (1.1) and setting N = Kx in Lemma (1.2) in [K]). We would like to mention also the following result in this area due to Ein and Lazarsfeld [EL]: For a big and numerically effective divisor L on a smooth complex projective threefold X, if L-C > 3 and L 2 'S > 7 for any curve C in X and any surface S in X, then L+Kx is generated by global holomorphic sections on X. An earlier version of this paper gave the following bound for Theorem (0.1) : ml is very ample for m > (24^(7(1 + C) 71 )^6713 ) 71 with C = ({n + 2)L + J^)!/ 1 " 1, which is by several order of magnitude not as sharp as the bound stated in the present Theorem (0.1). The earlier version of Theorem (0.2) gave the very ampleness of ml B for m > (24n n C(l + G) 71 ) 71^3) 71 with C = ((n + 2)L + B + Kx)^-^ After the earlier version was circulated, D ly told me a way to simplify the original proof and the simplication yielded the much sharper bound for Theorem (0.1). The simplification is to verify directly the numerical effectiveness of some line bundle by using its holomorphic sections on subvarieties of decreasing dimensions and to bypass the step, in the earlier version, of extending first those sections to the ambient manifold. I would like to thank D ly for the simplication of the proof and the sharpening of the bound. One lemma in the proof of our effective Matsusaka Big Theorem uses the strong Morse inequality of D ly which D ly obtained by analytic methods. Lawrence Ein and Robert Lazarsfeld told me a simple algebraic proof of that lemma which avoids the use of the strong Morse inequality of D ly. F. Catanese also has a simple algebraic proof of that lemma similar to the proof of Ein and Lazarsfeld. Both simple algebraic proofs of that lemma are reproduced here. I would like to thank Ein, Lazarsfeld, and Catanese for their simple algebraic proofs of that lemma. The method of proof of Theorem (0.2) uses the strong Morse inequality and the numerical criterion of very ampleness of D ly [Dl], [D2]. The earlier version used also NadePs vanishing theorem [N] in order to extend some holomorphic sections of a line bundle from a subvariety to the ambient manifold. Kollar's result (whose proof is algebraic geometric) can be used instead of the very ampleness criterion of D ly [D2] whose proof uses analysis. With the use of Kollar's result and the simple algebraic proofs of the lemma mentioned above due to Ein-Lazarsfeld and Catanese,

5 1390 YUM-TONG SIU the proof of our effective Matsusaka Big Theorem in this paper can be done purely algebraically, though the bound of Kollar's result is weaker than that of D ly and therefore results in a bound that is less sharp. The main idea of the proof of the effective Matsusaka Big Theorem in this paper is the following. Because of the very ampleness criterion of D ly and Kollar, it suffices to show that for a given numerically effective line bundle B and an ample line bundle L there is an effective lower bound m such that m L - B is numerically effective. This is done in three steps. The first step is a lemma on the existence of nontrivial holomorphic sections of a multiple of the difference of two ample line bundles whose Chern classes satisfy a certain inequality. This is the lemma for which the strong Morse inequality of D ly is used and for which Ein-Lazarsfeld and Catanese gave simple algebraic proofs. Such a nontrivial holomorphic section enables us to construct a closed positive current which is a curvature current of the line bundle and the curvature current will be used in an application of L 2 estimates of 9 to construct holomorphic sections. The second step is to produce, for any d-dimensional irreducible subvariety Y of X and any very ample line bundle H of X, a nontrivial holomorphic section over Y of the homomorphism sheaf from the sheaf of holomorphic d-forms of Y to (3A(A - 1)/2 - d - 1)H\Y for A > H^ V. The sheaf of holomorphic d-forms of Y is defined from the presheaf of holomorphic p-forms on the regular part of Y which are L 2. This second step is obtained by representing Y as a branched cover over a complex projective space and the section is obtained by constructing explicitly a global sheaf-homomorphism by solving linear equations by Cramer's rule. The key point is that since the canonical line bundle of the complex projective space of complex dimension d is equal to - (d + 1) times the hyperplane section line bundle, by representing a compact complex space as a branched cover of the complex projective space, one can relate its canonical line bundle to a very ample line bundle on it. The third step is to get the numerical effectiveness of ml B. The earlier version of this paper used NadePs vanishing theorem for multiplier ideal sheaves [N] to produce global holomorphic sections of high multiples of m L B so that the dimension of the common zero-sets of these sections is inductively reduced to zero. A new stratification of unreduced structure sheaves by multiplier ideal sheaves was introduced in the earlier version to carry out the induction process. D ly's simplication verifies the numerically effectiveness of ml - B by producing holomorphic sections of high multiples of m L - B on subvarieties of decreasing dimensions. The bypassing of the extension

6 EFFECTIVE MATUSAKA THEOREM 1391 of those holomorphic sections (called (7dj in 4) to the ambient manifold makes it unnecessary to use NadePs vanishing theorem for multiplier ideal sheaves and the new way of stratifying unreduced structure sheaves. Though not used here, such a new stratification of unreduced structure sheaves by multiplier ideal sheaves can be applied in other context, e.g. to get a D ly type very ampleness criterion for m L 4-2 Kx without using the analytic tools of the Monge-Ampere equation, but we will not discuss it here. 1. Effectiveness of multiples of differences of ample line bundles. First we state the strong Morse inequality due to D ly [Dl]. Let X be a compact complex manifold of complex dimension n, E be a holomorphic line bundle over X with smooth Hermitian metric along its fibers whose curvature form is 0{E)^ X{< q) be the open subset of X consisting of all points of X where the curvature form 6(E) has no more than q negative eigenvalues. D ly proved the following strong Morse inequality [Dl] : y^-l^dim^x, ke) < kn - f {-l) q (e{e)) n + o(k n ). J^O n ' ^(<9) Here o(a; 71 ) means the Landau symbol denoting a term whose quotient by k 71 goes to 0 as k > oo. We now apply D ly's strong Morse inequality to the case where E == F G with F and G being Hermitian holomorphic line bundles over a compact Kahler manifold X with semipositive curvature forms. LEMMA (1.1). Let F and G be holomorphic line bundles over a compact Kahler manifold X with Hermitian metrics so that their curvature forms are semipositive. Then for every 0 < q <^ dim X, ^ (-l^dim^(x,fc(f-g))<^ ^ (-I^MF^G^+O^) 0<,j<,q as k > oo. ' 0<j<q v / Proof. Let 0{F) and 0(G) be respectively the curvature forms of F and G which are smooth and semipositive. Let uj be the Kahler form of X. For e > 0, let (9e(F) = 0(F) + e uj and (9,(G) = 0(G) + e uj. Then

7 1392 YUM-TONG SIU 0(F - G) = 6^(F) - 0e(G). Let Ai >.. > An > 0 be the eigenvalues of 0e(G) with respect to 0e(F). Then X(< q) is precisely the set of all x in X such that Ag+i(a:) < 1. D ly's strong Morse inequality [Dl] now reads (I.I.I) E (-l)^dimjp(x,fc(f-g)) 0<J<q ^ I (- 1 ) 9 n ( 1 - ww + o(^). ^X^q) Kj<n Let crj^ be the j^ elementary symmetric function in Ai,, A^. We use the convention that a^ = 1 when j = 0 and a{ = 0 when j < 0. Then (1.1.2) Qwr^'w^ = ^w. We claim that E (- 1 ) 9-J < > (-i) 9 0<J'<9 n ( 1 -^) l<j<n for Ag+i < 1. We verify the claim by induction on n. Assume that it is true when n is replaced by n - 1. From it follows that <=<-^^n~-\\n ^ (-1)^<= ^ (-l^-^+a^) (Kj<g 0<j^g =^(-ir^_i-a» ^ (-ir^.i O^J'^g 0<j<g-l =(I-A«) ^ (-ir^-i+a^_, o$j<g >(I-A«) ^ (-ir^-i > (-i) 9 n ( 1 -^)- 0<j<g l<j<n This proves the claim. The lemma now follows from the claim and (I.I.I) and (1.1.2) when we let e ^ 0. Q In the earlier version of this paper the argument for the lemma was carried out only for q = 1 and was applied to get a nontrivial holomorphic section of k{f - G) over X for k sufficiently large under the assumption [nm. x (F-Gr-Y, Q n )F n - 2^>o, g=l v - I/

8 EFFECTIVE MATUSAKA THEOREM 1393 where the square bracket [ ] means the integral part. There the same method of simultaneous diagonalization of curvature forms was used but with less precise analysis of the inequality concerning the eigenvalues Ai,---,A^. This more precise form for a general q was suggested to me by D ly. In this paper the lemma will be applied only to the case q = 1 in the form of the following corollary. COROLLARY (1.2). Let X be a compact protective algebraic manifold of complex dimension n and F and G be numerically effective line bundles over X such that F 71 > nf n ~ l G. Then for k sufficiently large there exists a nontrivial holomorphic section ofk(f G) over X. Ein-Lazarsfeld and Catanese independently obtained similar algebraic geometric proofs of this corollary. We present them here. The proof of Ein-Lazarsfeld. For a coherent sheaf T over X, let /^(.F) denote dime I:P(X, F). By replacing F and G by F+L and G+L for some ample line bundle L over X and some sufficiently large we can assume without loss of generality that both F and G are ample. By multiplying both F and G by the same large positive number, we can assume without loss of generality that F and G are both very ample. Choose a smooth irreducible divisor D in the linear system G. Consider the exact sequence 0 -. Ox{k(F - G)) ^ Ox{kF) -^ OkD^kF) -> 0. It is enough to show that for k sufficiently large h (Ox{kF)) > h (OkD{k F)). There is a natural filtration of the sheaf 0^{k F) whose quotients are sheaves of the form OkD^k F j D) for 0 < j < k 1. So k-l h (OkD(kF))<^h (OkD{kF-jG)). 3=0 On the other hand, since G is very ample, we can choose a second smooth irreducible divisor D' in the linear system G which meets D in normal crossing. Since OkD (kf-j G) is isomorphic to OkD (kf-jd') which is a subsheaioiokd{kf), it follows that h {OkD(kF -j G)) <: h {OkD{kF)). Thus h (OkD(kF)) < k. /I (OD(A:F)). By Kodaira's vanishing theorem and the theorem of Riemann-Roch ^w,, t-^y ^, ^ ^,

9 1394 YUM-TONG SIU 1.71 pn whereas h {Ox(kF)) = + o(k n ), from which we obtain the conclun! sion. Catanese's proof. For a line bundle H over X, let h p (X, H) denote dime HP(X, H). As in the proof of Ein-Lazarsfeld we can assume without loss of generality that F and G are both very ample. Let k be any positive integer. We select k smooth members Gj, 1 < j < k in the linear system G and consider the exact sequence (1.2.1) 0 ^ Jf (x, kf-^gj) -^ H {X^ kf) -. Q)H (G^ kf\gj). 3 J=l By (1.2.1) and Kodaira's vanishing theorem and the theorem of Riemann- Roch ft (X, k{f - G)) > ^ - o^) - ^(-^ F-i. G, - o(k n - l )} n. ^ \^n L)\ / > kr -(F n -nf n - l 'G)-o(k n ). TL\ So for k sufficiently large there exists nontrivial holomorphic section of k(f - G) over X. What we need from Corollary (1.2) is a curvature current for the difference of the two numerically effective line bundles. Let L be a holomorphic line bundle on a compact complex manifold X of complex dimension n. Let X be covered by a finite open cover {Uj} so that L\Uj is trivial and let gjk on Uj D Uk be the transition function for L. A curvature current 0(L) for L is a closed positive (l,l)-current on X which is given by /^~1 99(pj on Uj where (pi is a plurisubharmoinc function on Uj such that ^^IPjfcl 2 = e"-^ on Uj n Uk. The collection {e-^} defines a (possibly nonsmooth) Hermitian metric along the fibers ofl. A closed positive (1,1)- current 0 means a current of type (1,1) which is closed and is positive in n-l the sense that 6 A ]~[ (v^to^ A aj) is a nonnegative measure on X for any j=i smooth (l,0)-forms a^(l <, j ^ n- 1) with compact support. A closed positive (l,l)-current is characterized by the fact that locally it is of the form V^IQQ^ for some plurisubharmonic function ^. The Leiong number, at a point P, of a closed positive (l,l)-current 0 on some open subset of C 71 is defined as the limit of (^r 2 Y~ n e^ ( ^-^l^l 2 ) 71 1 over the ball of radius r centered at P as r ^ 0, where z = (zi,, Zn) is the coordinate of C".

10 EFFECTIVE MATUSAKA THEOREM 1395 The normalization is such that the Leiong number of 99\og \z\ 2 is 1 27T at 0. The Leiong number is independent of the choice of local coordinates. The function e~^ is locally integrable at P if the closed positive (1,1)- current 99^ has Leiong number less than 1 at P. The function e~^ 27T is not locally integrable at P if the closed positive (l,l)-current 99i^ 2?r has Leiong number at least n at P. See e.g. [L] and [S] for more detailed information on closed positive currents and Leiong numbers. COROLLARY (1.3). Let X be a compact projective algebraic manifold of complex dimension n and F and G be numerically effective line bundles over X such that P 71 > nf n ~ l G. Then there exists a closed positive current which is the curvature current for F G. Proof. Let X be covered by a finite open cover {Uj} so that {F G)\Uj is trivial and let g^ on Uj D Uf. be the transition function for F G. Let s be a nontrivial holomorphic section for k(f G) for some sufficiently large k. The section s is given by a collection {sj} where sj is a holomorphic function on Uj with sj = g^s^ on Uj D U^. Define ^ = - log \Sj\ 2. Then the closed positive current on X defined by K 99(pj on Uj is a curvature current for F G. 27T D 2. The use of the very ampleness criterion of D ly. In 1 we obtain nontrivial sections for high multiples of the difference of ample line bundles, but we cannot control yet the effective bound for the multiple. In this section we are going to use closed positive currents and the very ampleness criterion of D ly [D2] to get an effective bound for the multiple. The very ampleness criterion of D ly [D2] states that for any ample line bundle L over a compact complex manifold X of complex dimension n, the line bundle lin^l-^-ikx is a very ample line bundle over X. We will use the more general result of D ly on the global generation of s-jets to further improve on the bound. The argument works with the very ampleness of 12n n L-\-2Kx but with a slightly worse bound. The result of D ly on the global generation of 5-jets states that, if L is an ample line bundle over a compact complex manifold X of complex dimension n,

11 1396 YUM-TONG SIU the global holomorphic sections of m L + 2Kx generate the s-jets at every point of X if m ^ 6(n + sy. Let X be a compact projective algebraic manifold and Y be an irreducible subvariety of complex dimension d in X. Let {jjy be the sheaf on Y defined by the presheaf which assigns to an open subset U of Y the space of all holomorphic d-forms on U n Reg Y which are L 2 on U D Reg V, where Reg Y is the set of all regular points of V. The sheaf ujy is coherent and is equal to the zeroth direct image of the sheaf of holomorphic d-forms on a desingularization Y' of Y under the desingularization map Y' >' Y. PROPOSITION (2.1). Let L be an ample line bundle over a compact complex manifold X of complex dimension n and let B be a numerically effective line bundle over X. Let Y be an irreducible subvariety of complex dimension d in X. Let Cn = 3(3n 2) n. Then for m > d T d i. R. y there exists a nontrivial holomorphic section of the sheaf o/y 0 0{mL - B + Kx + CnL)\Y over Y. Proof. Let TT : V > Y be a desingularization of V. Let E be an ample line bundle of Y'. We apply Corollary (1.3) to the numerically effective line bundles F = A;7r*(mL) and G = E + k^b over Y' for some sufficiently large k. The choice of m implies that F d > df^1 ' G for k sufficiently large. Hence there exists a closed positive (l,l)-current 6 on V which is a curvature current for k(f G) E. Choose a point P in Y' such that 7r(P) is in the regular part of Y and the Leiong number of 0 at P is zero. Since the global holomorphic sections of p L + 2 Kx over X generate the 2d-jets at every point ofx, for p > 6(n+2d) 71, there exists a Hermitian metric along the fibers of pl + 2 Kx which is smooth on X 7r(P) and whose Leiong number at 7r(P) is 2d. Thus for p > 3(n + 2d) n there exists a Hermitian metric along the fibers oipl-\-kx which is smooth on X 7r(P) and whose Leiong number at 7r(P) is d. Since d < n 1, we can use p = Cn' We give E a smooth metric with positive definite curvature form on V. By putting together the nonsmooth Hermitian metrics of k(cnl+kx) and k(f G) E and the smooth Hermitian metric of E, we get a nonsmooth Hermitian metric h of 7r*(mL B 4- Kx + CnL) such that the curvature current 0 of h has Leiong number d at P and has Leiong number zero at every point of U {P} for some open neighborhood U of P in V. Moreover, the curvature current 0 is no less than times the positive k definite curvature form of E on Y. We can assume without loss of generality

12 EFFECTIVE MATUSAKA THEOREM 1397 that both line bundles 7r*(mL - B + Kx 4- CnL)?7 and KY'\U are trivial. Let a be a holomorphic section of 7r*(mL - B + ^x + CnL) + J^y/ over / such that a is nonzero at P. We give Ky any smooth metric so that from h we have some nonsmooth metric h' of ^{ml B + JQc + CnL) + JCy/. We take a smooth function p with compact support on U so that p is identically 1 on some open neighborhood TV of P in U. Since the (7r*(mL B + Kx + CnL) 4- JCy/)-valued (9-closed (0,l)-form (9p)a on X which is supported on U is zero on iy and since the Leiong number of the curvature current of the metric h' is 0 at every point of U {P}, it follows that (9p)a is L 2 with respect to the metric h 1 of 7r*(mL B + Kx + CnL) + J<Ty/. Since the curvature current 6 of the line bundle 7r*(?nL B + -K^ + CnL) is no less than some positive definite smooth (l,l)-form on V, by the L 2 estimates of 9 (see e.g. [D2, p.332, Prop.4.1]) there exists some L 2 section r of7r*(ml-b+^x+cn^)+^y/ over V such that 9r = (9p)a on Y 1 '. Since the Leiong number of 0 is d at P, it follows that r vanishes at P and p a r is a global holomorphic section of7r*(ml i?+.k^+cn-l)+-ky/ over Y' which is nonzero at P. The direct image of p a r with respect to TT : Y' > Y is a global holomorphic section of o;y (g) 0(mL B + Kx + CnL) y over V which is non identically zero. D 3. Sections of the sum of a very ample line bundle and the anticanonical bundle. Because of the very nature of the very ampleness criterion of D ly, in 2 we could only construct holomorphic sections of line bundles containing the canonical line bundle as a summand. In this section we are going to construct holomorphic sections of the sum of a very ample line bundle and the anticanonical line bundle which will later be used to obtain holomorphic sections of line bundles without the canonical line bundle as a summand. The idea is that since the canonical line bundle of the complex projective space of complex dimension d is equal to (d 4-1) times the hyperplane section line bundle, by representing a compact complex space as a branched cover of the complex projective space, one can relate its canonical line bundle to a very ample line bundle on it. Let Y be a proper irreducible d-dimensional subvariety in P^y and HN be the hyperplane section line bundle of PN- Let A be a positive number no less than the degree H^ 'Y of Y. We use the notation introduced in 2 for ujy which is the sheaf on Y from the presheaf of local holomorphic

13 1398 YUM-TONG SIU d-forms on the regular part of Y which are L 2 locally at the singular points ofv. Let V be a linear subspace of P^y of complex dimension N d 1 which is disjoint from Y and let W be a linear subspace of P^y of complex dimension d which is disjoint from V. Denote by TT' the projection map from PN V to W defined by the linear subspaces V and W of P^y, which means that for x e PN V the point ^(x) is the point of intersection of W and the linear span of x and V. We identify W with Pd and let TT : Y ) Pd = W be the restriction of TT' to Y. We denote the restriction of HN to Pd = W by Hd so that Hd is the hyperplane section line bundle of Pd' We denote the restriction of HN to Y by H. LEMMA (3.1). There exists a nontrivial holomorphic section of the sheaf over K Hom(ujY, Oy((3A(A - 1)/2 - d - 1)H) Proof. Since there are nontrivial holomorphic sections of any positive multiple of H over V, without loss of generality we can assume that A is equal to the degree H^ Y of Y. Let Z C Pd be the branching locus oftt-.y ^Pd. Let D be some hyperplane in Pd not contained entirely in Z. Let D* be the hyperplane in P^y containing D and V so that D* is the topological closure in P^y of 7^/~ 1 (D). Let SD* be the holomorphic section of HN over PN whose divisor is D*. Let / be a holomorphic section of HN over PN such that h := assumes A distinct values at the A points of SD* ^~ l (Pl) for some Pi in P^. Let SD be the holomorphic section of Hd over Pd whose divisor is D. Let zi,, Zd be the inhomogeneous coordinates of Pd D. Then t = (soy^dz-i A A dzd can be regarded as a nowhere zero (d+ 1)[^] valued holomorphic d-form on Pd. The nontrivial holomorphic section of 7^om(o;y, Oy(3(A(A 1)/2 d 1)H) over Y will be defined by using h and t. We will define it in the following way. We take a holomorphic section 0 of ujy over an open neighborhood U of some point P of Y and we will define a holomorphic section of Oy((3A(A 1)/2 d 1)H over an open neighborhood of P in U. Without loss of generality we can assume that U is of the form TT'^I/') for some open subset U' of Pd. We consider the Vandermonde determinant defined for the A values

14 EFFECTIVE MATUSAKA THEOREM 1399 of h\y on the same fiber of TT : Y > P^. In other words, we take P e Pd such that 7T~ 1 (P) has A distinct points P^,---,P^ and we form the determinant /lo=det((/l(p^)) l/ - l )l<^<a. Let /i' = (^o) 2 - Then /i' is a meromorphic function on Pd with pole only along D. By considering the pole order of h along Tr^D), we conclude A-l that the pole order of h 1 along D is 2 ^AA=A(A 1). We write ^=1 A-l (3.1.1) 0(P^) = ^ a,(p)(fc(p^))ni < j < A) 1^=0 for some d-form dy on U'(Q < v < A 1). We solve the system (3.1.1) of equations by Cramer's rule. Let c^} = h{p^y~ 1 for p -^ v and c^ = O(P^). Let by be the determinant of the A x A matrix (c^)i<p,p,^\- Then a^(p) = b^l h'q = b^h'^l h'. We are going to use the removable singularity for L 2 holomorphic top degree forms so that an L 2 holomorphic d-form on fl. Z for some open subset f^ of Pd extends to a holomorphic d-form on Q. Thus, since 0 is an L 2 holomorphic d-form on U D RegY, we conclude from the pole orders of h and h' that the restriction of s^{x ~ l)/<2 h f a^ to U' - {Z H D) is a holomorphic (3(A(A- 1)/2)^-valued d-form on U' (Z D D) and therefore s^'~ // ' 21 h' a^ is a holomorphic (3(A(A l)/2)j:fd-valued d-form on U' by removability of singularity of codimension two. We have 6 = ^ (a^o7r){^\y). The section ^(s^'^^h^o / 7r*(t) i/=o of (3A(A 1)/2 d 1)H over U is holomorphic, because A-l ^^A(A-1)/2^,^ y ^^ ^ ^^^A(A-1)/2^, ^^ OTT^Y). 1^=0 The section of Uom(ujY, Oy((3A(A - 1)/2 - d - 1)H) over Y to be constructed is now given by the map which sends the holomorphic section 6 of UJY over U to the holomorphic section Tr*^^"^72 /^ / 7r*(^) of (3A(A- 1)/2 - d - 1)H over U. D PROPOSITION (3.2). Let L be an ample line bundle over a compact complex manifold X of complex dimension n and let B be a numerically effective line bundle over X. Let Y be a proper irreducible subvariety of complex dimension d in X. Let H be a very ample line T d i. R. y bundle ofy. Let Cn = 3(3n - 2) 71. Then for m > d ' ' there

15 1400 YUM-TONG SIU exists a nontrivial holomorphic section of the holomorphic line bundle (3A(A - 1)/2 - d - 1)H + {ml - B + Kx + C^L) y over V. Proof. The proposition follows from Lemma (3.1) and Proposition (2.1) after we embed Y into a complex projective space by using the holomorphic sections of H over Y. D COROLLARY (3.3). Let L be an ample line bundle over a compact complex manifold X of complex dimension n and let B be a numerically effective line bundle over X. Let Y be a proper irreducible subvariety of complex dimension d in X. Let Cn = 3(3n - 2) 71 and H = 2 CnL + 2 Kx' j.d i. n. Y Then for m > d there exists a nontrivial holomorphic section of the holomorphic line bundle ((3A(A - 1)/2)H + ml - B)\Y over Y. Proof. The case d = 0 is trivial. Hence we can assume that d > 1 and n > 2. In that case 2Cn == 3(3n - 2)' 1 > 12 n^ and by D ly's very ampleness criterion the line bundle H is very ample over X. Since (d+ ^ ) H is very ample over X, the corollary follows from Proposition (3.2). We will need a statement similar to Corollary (3.3) for the case Y =X. PROPOSITION (3.4). Let L be an ample line bundle over a compact complex manifold X of complex dimension n >, 2 and let B be a numerically effective line bundle over X. Let Cn = 3(3n 2) 71 and rn l. D H = 2 CnL + 2Kx- Then for m > n there exists a nontrivial L^ holomorphic section of the sheaf ml B + H over X. Proof. By Corollary (1.3) there exists a nonsmooth Hermitian metric h\ for ml B whose curvature current 6\ is a closed positive (1,1)- current. Take a point P in X at which the Leiong number of 6\ is 0. By [D2] (see the proof of Theorem 11.6 on p. 364 and the proof of Corollary 2 on p. 369) there exists a nonsmooth Hermitian metric h^ for 2CnL + Kx such that P, (i) the Leiong number of the curvature current 0^ of h^ is at least n at (ii) the Leiong number of 0'z is < 1 at every point of U {P} for some open neighborhood U of P in X, (iii) ^2 1 s^ kss than some positive definite smooth (l,l)-form on X. D

16 EFFECTIVE MATUSAKA THEOREM 1401 Now we put together the metrics h\ and h^ to form a nonsmooth metric ^3 for ml B + 2CnL + Kx- Without loss of generality we can assume that at every point of U the Leiong number of 6\ is zero and that the holomorphic line bundles (ml B + 2CnL)\U and Kx\U are trivial. We give Kx any smooth metric so that from h we have some nonsmooth metric h' of ml B + H. We take a smooth function p with compact support on U so that p is identically 1 on some open neighborhood W of P in U. Since the Leiong number of 6\ + 0^ is < 1 on U {P}, it follows that the {ml - B + ^-valued 9-closed (0,l)-form (9p)a on X is L 2. By applying the L 2 estimates of 9, we obtain an L 2 section T of ml B -{- H over X such that Qr = {Qp)(J on X. Since the Leiong number of 0i + ^2 is at least n at P, it follows that r vanishes at P and p cr r is a global holomorphic section of ml B + H over X which is nonzero at P. D 4. Final step in the proof of the effective Matsusaka Big Theorem. Since the effective Matsusaka Big Theorem is obviously true for the case when X is of complex dimension of one, we can assume that the complex dimension n of X is at least 2. To get Matsusaka 5 s Big Theorem, it is enough to get an effective bound on m for m L (B + 2Kx +pl) to be numerically effective for some p > 2(n+l), because then by D ly's very ampleness criterion [D2, p. 370, Remark 12.7] the holomorphic line bundle {12n n -^m-p)l-b = 1271^+2 Kx +m L- {B+2Kx -^-pl) becomes very ample. We use B + 2Kx + pl instead of just B + 2Kx, because we need Fujita's result [F] on the numerical effectiveness of (n + 1)L + Kx- The numerical effectiveness of m L {B + 2Kx + pl) is verified by evaluating it on an arbitrary compact curve. The main idea is to use Corollary (3.3). We can obtain nontrivial holomorphic sections of the line bundle of the form m L B over subvarieties and inductively we apply the argument to subvarieties which are the zero sets of such nontrivial holomorphic sections. This way we will get a numerical criterion for the numerical effectiveness of m L B and then afterwards we will replace B by B + 2Kx + pl. Let Cn = 3(3n - 2) 71 and H = 2CnL +2 Kx. As in the outline of the argument in the preceding paragraph, to obtain the numerical effectiveness of m L B, we are going to use Corollary (3.3) to construct inductively a sequence of (not necessarily irreducible) algebraic subvarieties

17 1402 YUM-TONG SIU X = Yn D Vn-i D D YI D Yo and positive integers m^ (1 < d < n) with the following two properties : (i) Yd is d-dimensional. (ii) For every irreducible component Ydj of Yd, there exists a nontrivial holomorphic section a^ of m^l - B over Ydj such that Yd-i is the union of the zero-sets of adj when j runs through the set indexing the branches YdjofYd. To start the induction, by Proposition (3.4) we can use m. > L^^B+H) _ n Jn and S^ a nontrivial holomorphic section On of m^l - B over X. Let V^_i be the zero-set of cr^. Suppose we have constructed Yy for d < v < n and my for d < ^/ < n. We are going to construct Yd-i and md. Let C, = 2(C, - (n + 1)). Then H - C, L = 2((n + 1)L + ^) is numerically effective over X by Fujita's result [F]. To choose m^ we have to compute Ydj ' H d and YdjL^B for every irreducible component Ydj of Yd. Now Ydj is a branch of the zero-set of some nontrivial holomorphic section 0^4.1^ of rrid+il - B over Vd+i^ for some branch Vd+i^ of Yd. For any numerically effective line bundles '1,, Ed over X, we have Ydj ;i. Ed < Yd^k (rrid^il- B). E^.. Ed < Yd+i,k rrid^il E^ ' Ed <y^. 77^.^...^ ^n (by the numerical effectiveness of H - C^ L) which by induction on d yields m^ V-7 JLd^ ^l E^. ^ri F^ S < mn -^- md H + l rjn-d r Ai v ^. Hence ^n YdJHd < c' ^n c' V/7 ^d ^d^tl S < mn -^7- ^^l H Tjn, ^n ^n Yd^B^^^^H^L^B. ^n ^n To apply Corollary (3.3) we impose on m^ (1 < d < n) the condition that md > L^Yd~ ( Ld-l ' YdJ ' ( B + 1^^ ~ 1)H )) for some ^ ^ ^^ ^d for all j. Since L^ Ydj >. 1 o^d H - C^ L is numerically effective, to get the condition we can let mn md^ n d - "C 1 ' " ' T^' 11 ^n ^n and set m, > -^^( Hn^ B + J^(^ _ ^ ^ ^er to get a simpler closed expression later we replace the last inequality by the stronger

18 EFFECTIVE MATUSAKA THEOREM 1403 inequality that m^ is no less than the integral part of y,\3 d / tin l. D ( +. ). Once these inequalities are satisfied, by Corollary (3.3) we can find a nontrivial holomorphic section adj of m^l B over Ydj. Our y^-i is now the union of the zero-sets of all adj when j runs through the set indexing the branches Ydj of Yd. The above inequality for md (1 <, d < n) is satisfied if inductively md is no less than the integral part of n rh^'b 3\ /m^ rnd^i \ 3 (C^\ H- '2AG, '" C^11 ) ' To get a closed formula, we consider for 1 <_ d < n the equations, n (H-^B 3wgn gd+i ^3 (W^ ^n 2A^"'C^ Y * Then ^ = 77^2^+1 and ^ = ^^4^-^-2^"' for 1 < d < n. Thus we \^n) \^n) can inductively define md to be the integral part of ((g^t(g") 2 (g"- l B+jg")) 4 "" d^(n(gn)2(gn-l.g+jgn))4 TC - i (cy-% ~ w-^ Clearly md < m\ for 1 < d < n. We now verify that the line bundle m\l B is numerically effective over X. Let F be an irreducible complex curve in X. We have to verify that {m-^l B) -T is nonnegative. There exists an integer 1 < d <: n such that F is contained entirely in some branch Ydj of Yd but is not contained entirely in Yd-i. There exists a nontrivial holomorphic section adj of m^l B over Ydj whose zero-set is contained in Yd-i. Thus adj does not vanish identically on F and we conclude that (m^l B) ' F is nonnegative. Since (mi md)l is ample, it follows that {m\l B) - F is nonnegative. This completes the proof that m\l B is numerically effective. Since 2Cn > ^r^ for n > 2, by [D2, p. 370, Remark 12.7] from the numerical effectiveness of m\l B it follows that the line bundle m\l B + H is very ample. After replacing B by B + H^ we conclude that m L B is very ample for and Theorem (0.2) is proved. {n(h n ) 2 (H n - l 'B+^H n )^n~ l m ^> ~ 3 (C'n) 471 " 171 " 1 We now derive Theorem (0.1) from Theorem (0.2) by using the following inequalities of Chern numbers of numerically effective line bundles

19 1404 YUM-TONG SIU [D2, Prop. 5.2(b)]. If L^'.,Z^ are numerically effective holomorphic line bundles over a compact projective algebraic manifold X of complex dimension n and k^''., ke are positive integers with k^ + + ke = n, then L^... L^ > (L^^... (Z^)^. We now apply the inequality to the case of Li = F + G and L^ = F. Then or (F + G')F 71-1 > ((F + G?) 71 ) 1 /^71^71-1 )/^ (F^Gr< ( - (F^G)Fn^ v / - (^n)n-l Let F = {Cn - (n+ 1))L and G = (n+ l)l+^x. Then F+G = l^ and Thus A^n^ ((Gn^+^x)^71-1 )" 2^ - (L 71 ) 71-1 (n 5^) 3 ) 471 " 1 < (Q3n-l^)4^/((G^+^). 7.^)^3.4^ / \ 2' / - v / V (L 71 ) In the case of B = 0 the line bundle m L is very ample if ^ ^ (2 3n - 1 5n) 4T^ l ((G^+^)^n- l 13n ) 4T^" Thus Theorem (0.1) is proved. - (^4--in-J^4-i3(n-l) BIBLIOGRAPHY [Dl] [D2] [EL] [F] [K] J.-P. D LY, Champs magnetiques et inegalites de Morse pour la d" cohomologie, Compte-Rendus Acad. Sci, Serie I, 301 (1985), and Ann Inst Fourier, 35-4 (1985), J.-P. D LY, A numerical criterion for very ample line bundles J Diff Geom 37 (1993), L. EIN and R. LAZARSFELD, Global generation of pluricanonical and adjoint linear series on smooth projective threefolds, preprint, T. FUJITA, On polarized manifolds whose adjoint bundles are not semipositive, Proceedings of the 1985 Sendai Conference on Algebraic Geometry, Advanced Studies in Pure Mathematics, 10 (1987), J. KOLLAR, Effective base point freeness, Math. Ann., to appear. [KM] J. KOLLAR and T. MATSUSAKA, Riemann-Roch type inequalities Amer J Math., 105 (1983), [L] P. LELONG, Plurisubharmonic functions and positive differential forms, Gordon and Breach, New York, 1969.

20 EFFECTIVE MATUSAKA THEOREM 1405 [LM] D. LIEBERMAN and D. MUMFORD, Matsusaka's Big Theorem (Algebraic Geometry, Arcata 1974), Proceedings of Symposia in Pure Math., 29 (1975), [Ml] T. MATSUSAKA, On canonically polarized varieties II, Amer. J. Math., 92 (1970), [M2] T. MATSUSAKA, Polarized varieties with a given Hilbert polynomial, Amer. J. Math., 94 (1972), [N] [S] A. NADEL, Multiplier ideal sheaves and existence of Kahler-Einstein metrics of positive scalar curvature, Proc. Nat. Acad. Sci. U.S.A., 86 (1989), and Ann. of Math., 132 (1990), Y.-T. SlU, Analyticity of sets associated to Leiong numbers and the extension of closed positive currents, Invent. Math., 27 (1974), Yum-Tong SIU, Department of Mathematics Harvard University Cambridge, MA (U.S.A.).